JP4293005B2 - Speech and music signal encoding apparatus and decoding apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To excellently encode a sound music signal over the entire band by a voice music signal encoding and decoding device of band-division constitution. <P>SOLUTION: A residue vector is generated from a difference vector outputted from a first differentiator 180 by using a reverse filter 230. A band selecting circuit 250 generates (n) sub-vectors by using components included in an arbitrary band with an orthogonally-converted residue vector. An orthogonal conversion coefficient quantizing circuit 260 quantizes the (n) sub-vectors. <P>COPYRIGHT: (C)2004,JPO

Description

The present invention relates to an encoding device and a decoding device for transmitting a voice music signal at a low bit rate.

As a method of encoding an audio signal at a medium to low bit rate with high efficiency, a method of encoding an audio signal by separating it into a linear prediction filter and its driving excitation signal (excitation signal) is widely used.

One typical method is CELP (Code Excited Linear Prediction). In CELP, a linear prediction filter in which a linear prediction coefficient obtained by linear prediction analysis of input speech is set is driven by a sound source signal represented by the sum of a speech pitch period signal and a noisy signal. Thus, a synthesized voice signal (reproduced signal) is obtained. Regarding CELP, reference can be made to “Code excited linear prediction: High quality speech at very low bit rates” (Proc. ICASSP, pp. 937-940, 1985) (Non-Patent Document 1). Also, the coding performance for the music signal can be improved by adopting a band division configuration for the CELP. In this configuration, a reproduction signal is generated by driving a linear prediction synthesis filter with an excitation signal obtained by adding sound source signals corresponding to each band.

Regarding CELP in band division configuration, “Multi-band CELP Coding of Speech and Music” by A. Ubale et al. (IEEE Workshop on Speech Coding for Telec
ommunications, pp. 101-102, 1997) (Non-Patent Document 2).

FIG. 31 is a block diagram showing an example of a conventional speech and music signal encoding apparatus. Here, for simplicity, the number of bands is 2. An input signal (input vector) generated by sampling a voice or music signal and combining the plurality of samples into one vector as one frame is input from the input terminal 10.

The linear prediction coefficient calculation circuit 170 receives an input vector from the input terminal 10, performs linear prediction analysis on the input vector, obtains a linear prediction coefficient, further quantizes the linear prediction coefficient, and produces a quantized linear prediction coefficient. Ask for. The linear prediction coefficient is output to the weighting filter 140 and the weighting filter 141, and the index corresponding to the quantized linear prediction coefficient is output to the linear prediction synthesis filter 130, the linear prediction synthesis filter 131, and the code output circuit 190.

The first sound source generation circuit 110 receives an index output from the first minimization circuit 150, reads a first sound source vector corresponding to the index from a table storing a plurality of sound source vectors, Output to the first gain circuit 160.

The second sound source generation circuit 111 receives the index output from the second minimization circuit 151, reads the second sound source vector corresponding to the index from a table storing a plurality of sound source vectors, Output to the second gain circuit 161.

The first gain circuit 160 receives the index output from the first minimizing circuit 150 and the first sound source vector output from the first sound source generation circuit 110, and receives the first gain corresponding to the index. The gain is read from a table in which a plurality of gain values are stored, the first gain and the first excitation vector are multiplied, a third excitation vector is generated, and the third excitation vector is 1 to the band pass filter 120.

The second gain circuit 161 receives the index output from the second minimization circuit 151 and the second sound source vector output from the second sound source generation circuit 111, and receives a second corresponding to the index. The gain is read from a table in which a plurality of gain values are stored, the second gain and the second excitation vector are multiplied, a fourth excitation vector is generated, and the fourth excitation vector is 2 to the band-pass filter 121.

The first band pass filter 120 receives the third sound source vector output from the first gain circuit 160. The third sound source vector is band-limited to the first band by this filter to obtain a first excitation vector. The first band pass filter 120 outputs the first excitation vector to the linear prediction synthesis filter 130.

The second band pass filter 121 receives the fourth sound source vector output from the second gain circuit 161. The fourth excitation vector is band-limited to the second band by this filter to obtain a second excitation vector. The second band pass filter 121 outputs the second excitation vector to the linear prediction synthesis filter 131.

The linear prediction synthesis filter 130 receives the first excitation vector output from the first bandpass filter 120 and the index corresponding to the quantized linear prediction coefficient output from the linear prediction coefficient calculation circuit 170, and the index By reading the quantized linear prediction coefficient corresponding to the above from a table in which a plurality of quantized linear prediction coefficients are stored, and driving the filter in which the quantized linear prediction coefficient is set by the first excitation vector, A first reproduction signal (reproduction vector) is obtained. Then, the first reproduction vector is output to the first subtractor 180.

The linear prediction synthesis filter 131 receives the second excitation vector output from the second bandpass filter 121 and the index corresponding to the quantized linear prediction coefficient output from the linear prediction coefficient calculation circuit 170, and the index By reading the quantized linear prediction coefficient corresponding to the above from a table in which a plurality of quantized linear prediction coefficients are stored, and driving the filter in which the quantized linear prediction coefficient is set by the second excitation vector, A second reproduction vector is obtained. Then, the second reproduction vector is output to the second subtractor 181.

The first differentiator 180 inputs an input vector via the input terminal 10, inputs a first reproduction vector output from the linear prediction synthesis filter 130, calculates a difference between them, and calculates the first difference. The difference vector is output to the weighting filter 140 and the second differencer 181.

The second subtractor 181 receives the first difference vector from the first subtractor 180, receives the second reproduction vector output from the linear prediction synthesis filter 131, calculates the difference between them, Is output to the weighting filter 141 as the second difference vector.

The weighting filter 140 receives the first difference vector output from the first differentiator 180 and the linear prediction coefficient output from the linear prediction coefficient calculation circuit 170, and uses the linear prediction coefficient to human hearing. A weighting filter corresponding to the characteristic is generated, and the weighting filter is driven by the first difference vector to obtain a first weighting difference vector. The first weighted difference vector is output to the first minimizing circuit 150.

The weighting filter 141 receives the second difference vector output from the second differentiator 181 and the linear prediction coefficient output from the linear prediction coefficient calculation circuit 170, and uses the linear prediction coefficient to detect human hearing. A weighting filter corresponding to the characteristic is generated, and the weighting filter is driven by the second difference vector to obtain a second weighting difference vector. Then, the second weighted difference vector is output to the second minimizing circuit 151.

The first minimizing circuit 150 sequentially outputs indexes corresponding to all the first sound source vectors stored in the first sound source generating circuit 110 to the first sound source generating circuit 110 to obtain a first gain. Indexes corresponding to all the first gains stored in the circuit 160 are sequentially output to the first gain circuit 160. Further, the first weighted difference vector output from the weighting filter 140 is sequentially input, its norm is calculated, and the first sound source vector and the first gain that minimize the norm are obtained. The selected indexes are output to the code output circuit 190.

The second minimizing circuit 151 sequentially outputs indexes corresponding to all the second sound source vectors stored in the second sound source generating circuit 111 to the second sound source generating circuit 111 to obtain a second gain. Indexes corresponding to all the second gains stored in the circuit 161 are sequentially output to the second gain circuit 161. Further, the second weighted difference vector output from the weighting filter 141 is sequentially input, the norm thereof is calculated, and the second sound source vector and the second gain that minimize the norm are obtained. The selected indexes are output to the code output circuit 190.

The code output circuit 190 inputs an index corresponding to the quantized linear prediction coefficient output from the linear prediction coefficient calculation circuit 170. Also, the second sound source output from the second minimizing circuit 151 by inputting an index corresponding to each of the first sound source vector and the first gain output from the first minimizing circuit 150. An index corresponding to each of the vector and the second gain is input. Each index is converted into a bit-sequence code and output via the output terminal 20.

FIG. 32 is a block diagram showing an example of a conventional speech and music signal decoding apparatus. A bit sequence code is input from the input terminal 30.

The code input circuit 310 converts the code of the bit series input from the input terminal 30 into an index. The index corresponding to the first sound source vector is output to the first sound source generation circuit 110. The index corresponding to the second sound source vector is output to the second sound source generation circuit 111. The index corresponding to the first gain is output to the first gain circuit 160. The index corresponding to the second gain is output to the second gain circuit 161. The index corresponding to the quantized linear prediction coefficient is output to the linear prediction synthesis filter 130 and the linear prediction synthesis filter 131.

The first excitation generator 110 receives the index output from the code input circuit 310, reads the first excitation vector corresponding to the index from a table storing a plurality of excitation vectors, Output to the gain circuit 160.

The second sound source generation circuit 111 receives the index output from the code input circuit 310, reads a second sound source vector corresponding to the index from a table storing a plurality of sound source vectors, Output to the gain circuit 161.

The first gain circuit 160 receives the index output from the code input circuit 310 and the first sound source vector output from the first sound source generation circuit 110, and obtains a first gain corresponding to the index, Reading from a table storing a plurality of gain values, multiplying the first gain and the first sound source vector to generate a third sound source vector, and using the third sound source vector as the first band Output to the pass filter 120.

The second gain circuit 161 receives the index output from the code input circuit 310 and the second sound source vector output from the second sound source generation circuit 111, and obtains the second gain corresponding to the index, Reading from a table storing a plurality of gain values, multiplying the second gain and the second sound source vector to generate a fourth sound source vector, and using the fourth sound source vector as the second band Output to the pass filter 121.

The first band pass filter 120 receives the third sound source vector output from the first gain circuit 160. The third sound source vector is band-limited to the first band by this filter to obtain a first excitation vector. The first band pass filter 120 outputs the first excitation vector to the linear prediction synthesis filter 130.

The second band pass filter 121 receives the fourth sound source vector output from the second gain circuit 161. The fourth excitation vector is band-limited to the second band by this filter to obtain a second excitation vector. The second band pass filter 121 outputs the second excitation vector to the linear prediction synthesis filter 131.

The linear prediction synthesis filter 130 receives the first excitation vector output from the first bandpass filter 120 and the index corresponding to the quantized linear prediction coefficient output from the code input circuit 310, and corresponds to the index. The first linear quantization coefficient is read from a table in which a plurality of quantized linear prediction coefficients are stored, and the filter in which the quantized linear prediction coefficient is set is driven by the first excitation vector. Get the playback vector. Then, the first reproduction vector is output to the adder 182.

The linear prediction synthesis filter 131 inputs the second excitation vector output from the second bandpass filter 121 and the index corresponding to the quantized linear prediction coefficient output from the code input circuit 310, and corresponds to the index. The quantized linear prediction coefficient to be read is read from a table in which a plurality of quantized linear prediction coefficients are stored, and the filter in which the quantized linear prediction coefficient is set is driven by the second excitation vector, so that the second Get the playback vector. Then, the second reproduction vector is output to the adder 182.

The adder 182 inputs the first reproduction vector output from the linear prediction synthesis filter 130 and the second reproduction vector output from the linear prediction synthesis filter 131, calculates the sum of these, Are output via the output terminal 40.
Code excited linear prediction: High quality speech at very low bit rates (Proc. ICASSP, pp.937-940, 1985) Multi-band CELP Coding of Speech and Music (IEEE Workshop on Speech Coding for Telecommunications, pp.101-102, 1997

The problem is that, in the above-described conventional speech and music signal encoding apparatus, an excitation signal having a band characteristic corresponding to the low frequency of the input signal and an excitation signal having a band characteristic corresponding to the high frequency of the input signal are added. Since the reproduction signal is generated by driving the linear prediction synthesis filter obtained from the input signal by the excitation signal obtained as described above, in order to perform encoding based on CELP in the band belonging to the high frequency range, A decrease in encoding performance in a band belonging to a high frequency range results in a deterioration in the encoding quality of a speech and music signal in all bands.

The reason for this is that signals in a band belonging to the high frequency range have properties that are significantly different from those of speech, so CELP modeling the speech generation process can accurately signal signals in the band belonging to the high frequency range. This is because it cannot be generated. An object of the present invention is to solve the above-described problems and provide a speech and music signal encoding apparatus that can satisfactorily encode speech and music signals over the entire band.

The first apparatus of the present invention generates a first reproduction signal by driving a linear prediction synthesis filter obtained from an input signal by an excitation signal corresponding to a first band, and the input signal and the first reproduction signal are generated. A residual signal is generated by driving an inverse filter of the linear prediction synthesis filter with a difference signal from the signal, and a component corresponding to the second band in the residual signal is encoded after orthogonal transformation.
Specifically, a unit (110, 160, 120, 130 in FIG. 1) for generating a first reproduction signal by driving the linear prediction synthesis filter with an excitation signal corresponding to the first band, and an input signal 1 to generate a residual signal by driving an inverse filter of the linear prediction synthesis filter using a difference signal between the first reproduced signal and the second reproduction signal, and a second signal in the residual signal. Means (240, 250, 260 in FIG. 1) for encoding the component corresponding to the band after orthogonal transformation.

The second device of the present invention generates the first and second reproduction signals by driving the linear prediction synthesis filter obtained from the input signal by the excitation signals corresponding to the first and second bands, A residual signal is generated by driving an inverse filter of the linear prediction synthesis filter using a difference signal between a signal obtained by adding the first and second reproduction signals and the input signal, and a third band in the residual signal is generated. The component corresponding to is encoded after orthogonal transformation. Specifically, means (1001, 1002 in FIG. 8) for generating the first and second reproduction signals by driving the linear prediction synthesis filter by the excitation signals corresponding to the first and second bands, A residual signal is generated by driving an inverse filter of the linear prediction synthesis filter using a difference signal between the input signal and a signal obtained by adding the first and second reproduction signals, and a third signal in the residual signal is generated. Means (1003 in FIG. 8) for encoding the component corresponding to the band after orthogonal transformation.

The third device of the present invention drives the linear prediction synthesis filter obtained from the input signal by the excitation signals corresponding to the first to (N-1) th bands, thereby generating the first to (N-1) th reproduced signals. Generating a residual signal by driving an inverse filter of the linear prediction synthesis filter with a difference signal between the input signal and a signal obtained by adding the first to (N-1) th reproduced signals, and generating the residual signal The component corresponding to the Nth band in the signal is encoded after orthogonal transformation. Specifically, means for generating the first to (N-1) th reproduced signals by driving the linear prediction synthesis filter with the excitation signals corresponding to the first to (N-1) th bands (1001, 1001 in FIG. 9). 1004) and a signal obtained by adding the first to (N-1) th reproduced signals and the input signal to drive an inverse filter of the linear prediction synthesis filter to generate a residual signal, and Means (1005 in FIG. 9) for encoding the component corresponding to the Nth band in the difference signal after orthogonal transformation.

The fourth device of the present invention, in the second encoding, drives the inverse filter of the linear prediction synthesis filter obtained from the input signal by the difference signal between the first encoded decoded signal and the input signal. A difference signal is generated, and a component corresponding to an arbitrary band in the residual signal is encoded after orthogonal transformation. Specifically, the means for calculating the difference between the first encoded decoded signal and the input signal (180 in FIG. 11) and the inverse filter of the linear prediction synthesis filter obtained from the input signal are driven by the difference signal. Means for generating a residual signal and encoding a component corresponding to an arbitrary band in the residual signal after orthogonal transformation (1002 in FIG. 11).

The fifth device of the present invention is the inverse of the linear prediction synthesis filter obtained from the input signal by the difference signal between the input signal and the signal obtained by adding the first and second encoded decoded signals in the third encoding. A residual signal is generated by driving a filter, and a component corresponding to an arbitrary band in the residual signal is encoded after orthogonal transformation. Specifically, means (1801, 1802 in FIG. 12) for calculating a difference signal between the signal obtained by adding the first and second encoded decoded signals and the input signal, and the linear prediction synthesis filter obtained from the input signal Means (1003 in FIG. 12) which generates a residual signal by driving an inverse filter with the differential signal and encodes a component corresponding to an arbitrary band in the residual signal after orthogonal transformation.

In a sixth apparatus of the present invention, a linear prediction synthesis filter obtained from an input signal based on a difference signal between an input signal and a signal obtained by adding the first to (N-1) th encoded decoding signals in the Nth encoding A residual signal is generated by driving the inverse filter, and a component corresponding to an arbitrary band in the residual signal is encoded after orthogonal transformation. Specifically, means (1801 and 1802 in FIG. 13) for calculating a difference signal between a signal obtained by adding the first to (N-1) th encoded decoding signals and the input signal, and linear prediction synthesis obtained from the input signal Means for generating a residual signal by driving an inverse filter of the filter with the differential signal, and encoding a component corresponding to an arbitrary band in the residual signal after orthogonal transformation (1005 in FIG. 13).

The seventh apparatus of the present invention uses the pitch prediction filter when generating the excitation signal corresponding to the first band of the input signal. Specifically, it has pitch prediction means (112, 162, 184, 510 in FIG. 14).

The eighth device of the present invention downsamples the first input signal sampled at the first sampling frequency to the second sampling frequency to generate a second input signal, and generates the second input signal from the second input signal. Driving a synthesis filter in which the obtained first linear prediction coefficient is set with an excitation signal to generate a first reproduction signal and up-sampling the first reproduction signal to the first sampling frequency And a second linear signal obtained by sampling frequency conversion of the linear prediction coefficient obtained from the first input signal and the first linear prediction coefficient into a first sampling frequency. Calculating a third linear prediction coefficient from the difference from the prediction coefficient, calculating a fourth linear prediction coefficient from the sum of the second linear prediction coefficient and the third linear prediction coefficient, A residual signal is generated by driving an inverse filter in which the fourth linear prediction coefficient is set by a difference signal between the input signal of 1 and the second reproduction signal, and is set to an arbitrary band in the residual signal. Corresponding components are encoded after orthogonal transformation. Specifically, the first input signal sampled at the first sampling frequency is down-sampled to the second sampling frequency to generate a second input signal (780 in FIG. 15), and the second Means for generating a first reproduction signal by driving a synthesis filter in which a first linear prediction coefficient obtained from the input signal is set by an excitation signal (770 and 132 in FIG. 15); Means (781 in FIG. 15) for generating a second reproduction signal by up-sampling the reproduction signal to the first sampling frequency, the linear prediction coefficient obtained from the first input signal, and the first linear Means for calculating the third linear prediction coefficient from the difference between the prediction coefficient and the second linear prediction coefficient obtained by converting the sampling frequency to the first sampling frequency (771 in FIG. 15) 772), a fourth linear prediction coefficient is calculated from the sum of the second linear prediction coefficient and the third linear prediction coefficient, and a difference signal between the first input signal and the second reproduced signal is calculated. Means for generating a residual signal by driving an inverse filter in which the fourth linear prediction coefficient is set (180, 730 in FIG. 15), and a component corresponding to an arbitrary band in the residual signal, Means for encoding after orthogonal transformation (240, 250, 260 in FIG. 15).

The ninth apparatus of the present invention generates an excitation signal corresponding to the second band by performing orthogonal inverse transform on the decoded orthogonal transform coefficient, and drives the linear prediction synthesis filter by the excitation signal to generate the second signal. Further, the first reproduction signal is generated by driving the linear prediction filter with the excitation signal corresponding to the decoded first band, and the first reproduction signal and the second reproduction signal are generated. Decoded speech music is generated by adding the reproduction signals. Specifically, a unit (440, 460 in FIG. 16) that generates an excitation signal corresponding to the second band by performing orthogonal inverse transform on the decoded signal and the orthogonal transform coefficient, and a linear prediction synthesis filter include the excitation signal. Means for generating a second reproduction signal by driving with (131 in FIG. 16) and means for generating the first reproduction signal by driving the linear prediction filter with an excitation signal corresponding to the first band (110, 120, 130, 160 in FIG. 16) and means (182 in FIG. 16) for generating decoded speech music by adding the first reproduction signal and the second reproduction signal.

The tenth device of the present invention generates an excitation signal corresponding to the third band by performing orthogonal inverse transform on the decoded orthogonal transform coefficient, and drives the linear prediction synthesis filter by the excitation signal to generate the third signal. And the first and second reproduction signals are generated by driving the linear prediction filter with the excitation signals corresponding to the decoded first and second bands, and the first to second reproduction signals are generated. 3 is added to the reproduced signal to generate decoded speech music. Specifically, an orthogonal inverse transform is performed on the decoded orthogonal transform coefficient to generate an excitation signal corresponding to the third band, and a linear prediction synthesis filter is driven by the excitation signal to thereby generate a third reproduction signal. Means for generating (1053 in FIG. 22) and means for generating the first and second reproduction signals by driving the linear prediction filter with excitation signals corresponding to the first and second bands (1051 in FIG. 22). , 1052) and means for generating decoded speech music by adding the first to third reproduction signals (1821, 1822 in FIG. 22).

The eleventh device of the present invention generates an excitation signal corresponding to the Nth band by performing orthogonal inverse transform on the decoded orthogonal transform coefficient, and drives the linear prediction synthesis filter by the excitation signal to generate the Nth filter. And the first to Nn−1 regenerated signals are generated by driving the linear prediction filter with the excitation signals corresponding to the decoded first to (N−1) th bands, Decoded speech music is generated by adding the first to Nth reproduction signals. Specifically, an orthogonal inverse transform is performed on the decoded orthogonal transform coefficient to generate an excitation signal corresponding to the Nth band, and a linear prediction synthesis filter is driven by the excitation signal to thereby convert the Nth reproduction signal. Means for generating (1055 in FIG. 23) and means for generating the first to (N−1) th reproduced signals by driving the linear prediction filter with the excitation signals corresponding to the first to (N−1) th bands. 23, 1051, 1054) in FIG. 23, and means (1821, 1822 in FIG. 23) for generating decoded speech music by adding the first to Nth reproduction signals.

The twelfth device of the present invention generates an excitation signal by performing orthogonal inverse transform on the decoded orthogonal transform coefficient in the second decoding, and driving a linear prediction synthesis filter with the excitation signal to generate a reproduction signal. Generating decoded speech music by adding the reproduction signal and the first decoded signal. Specifically, means (1052 in FIG. 24) generates an excitation signal by performing orthogonal inverse transform on the decoded orthogonal transform coefficient, and generates a reproduction signal by driving a linear prediction synthesis filter with the excitation signal. And means (182 in FIG. 24) for generating decoded speech music by adding the reproduction signal and the first decoded signal.

In the third decoding, the thirteenth device of the present invention generates an excitation signal by performing orthogonal inverse transform on the decoded orthogonal transform coefficient, and drives a linear prediction synthesis filter with the excitation signal to generate a reproduction signal. And generating the decoded speech music by adding the reproduction signal and the first and second decoded signals. Specifically, means for generating an excitation signal by orthogonally transforming the decoded orthogonal transform coefficient and generating a reproduction signal by driving a linear prediction synthesis filter with the excitation signal (1053 in FIG. 25); And means for generating decoded speech music by adding the reproduction signal and the first and second decoded signals (1821 and 1822 in FIG. 25).

The fourteenth device of the present invention generates an excitation signal by performing orthogonal inverse transform on the decoded orthogonal transform coefficient in the Nth decoding, and drives a linear prediction synthesis filter with the excitation signal to generate a reproduction signal. Generating decoded speech music by adding the reproduction signal and the first to (N-1) th decoded signals. Specifically, means (1055 in FIG. 26) for generating an excitation signal by performing orthogonal inverse transform on the decoded orthogonal transform coefficient and generating a reproduction signal by driving a linear prediction synthesis filter with the excitation signal. Means for generating decoded speech music by adding the reproduction signal and the first to (N-1) th decoded signals (1821 and 1822 in FIG. 26).

The fifteenth device of the present invention uses the filter relating to the pitch prediction filter when generating the excitation signal corresponding to the first band. Specifically, it has pitch prediction means (112, 162, 184, 510 in FIG. 27).

According to a sixteenth device of the present invention, a signal obtained by driving a first linear prediction synthesis filter with a first excitation signal for the first band is up-sampled to a first sampling frequency to obtain a first And a second excitation signal corresponding to the second band is generated by performing orthogonal inverse transform on the decoded orthogonal transform coefficient, and a second linear prediction synthesis filter is generated by the second excitation signal. The second reproduction signal is generated by driving and the decoded reproduction music is generated by adding the first reproduction signal and the second reproduction signal. Specifically, a signal obtained by driving the first linear prediction synthesis filter with the first excitation signal corresponding to the first band is up-sampled to the first sampling frequency to obtain the first reproduction signal. A second excitation signal corresponding to the second band is generated by performing orthogonal inverse transformation on the generated orthogonal transform coefficients (132 and 781 in FIG. 28) and the second excitation signal. Means for generating a second reproduction signal by driving the second linear prediction synthesis filter (440 and 831 in FIG. 28), and decoding by adding the first reproduction signal and the second reproduction signal Means for generating voice music (182 in FIG. 28).

The device of the present invention 17 decodes the code output from the device of the present invention 1 by the device of the present invention 9. Specifically, it has speech and music signal encoding means (FIG. 1) and speech and music signal decoding means (FIG. 16).

The apparatus of the present invention 18 decodes the code output from the apparatus of the present invention 2 by the apparatus of the present invention 10. Specifically, it has a voice music signal encoding means (FIG. 8) and a voice music signal decoding means (FIG. 22).

The device according to the nineteenth aspect of the present invention decodes the code output from the device according to the third aspect of the present invention by the device according to the present invention 11. Specifically, it has a voice music signal encoding means (FIG. 9) and a voice music signal decoding means (FIG. 23).

The device of the present invention 20 decodes the code output from the device of the present invention 4 by the device of the present invention 12. Specifically, it has a voice music signal encoding means (FIG. 11) and a voice music signal decoding means (FIG. 24).

The device of the present invention 21 decodes the code output from the device of the present invention 5 by the device of the present invention 13. Specifically, the audio / music signal encoding means (FIG. 12) and the audio / music signal decoding means (FIG. 25) are included.

The device of the present invention 22 decodes the code output from the device of the present invention 6 by the device of the present invention 14. Specifically, it has speech and music signal encoding means (FIG. 13) and speech and music signal decoding means (FIG. 26).

The device of the present invention 23 decodes the code output from the device of the present invention 7 by the device of the present invention 15. Specifically, it has a voice music signal encoding means (FIG. 14) and a voice music signal decoding means (FIG. 27).

The device of the present invention 24 decodes the code output from the device of the present invention 8 by the device of the present invention 16. Specifically, the audio / music signal encoding means (FIG. 15) and the audio / music signal decoding means (FIG. 28) are included.
(Function)
In the present invention, a first reproduction signal is generated by driving a linear prediction synthesis filter obtained from an input signal by an excitation signal having a band characteristic corresponding to a low frequency of the input signal, and the input signal and the first signal are generated. A residual signal is generated by driving an inverse filter of the linear prediction synthesis filter with a difference signal from the reproduction signal, and a high frequency component of the residual signal is encoded using an encoding method based on orthogonal transform. . That is, for a signal having a property different from that of speech in a band belonging to a high frequency range, encoding based on orthogonal transformation is performed instead of CELP. The coding based on the orthogonal transform has higher coding performance for signals having properties different from those of speech compared to CELP. For this reason, the encoding performance with respect to the high frequency component of the input signal is improved. As a result, it is possible to encode the audio music signal satisfactorily over the entire band.

The effect of the present invention is that the audio music signal can be encoded well over the entire band. The reason is that a first reproduction signal is generated by driving a linear prediction synthesis filter obtained from the input signal by a sound source signal having a band characteristic corresponding to a low frequency of the input signal, and the input signal and the first signal are generated. A residual signal is generated by driving an inverse filter of the linear prediction synthesis filter using a difference signal from the reproduced signal of the above-described reproduction signal, and a high frequency component of the residual signal is encoded using an encoding method based on orthogonal transform This is because the coding performance for the high frequency component of the input signal is improved.

FIG. 1 is a block diagram showing a configuration of a speech and music signal encoding apparatus according to a first embodiment of the present invention. Here, the number of bands is assumed to be two. An input signal (input vector) generated by sampling a voice or music signal and combining the plurality of samples into one vector as one frame is input from the input terminal 10.
The input vector is represented as x (n), n = 0,. However, L is a vector length. The input signal is band-limited from F _s0 [Hz] to F _e0 [Hz]. For example, assuming that the sampling frequency is 16 [kHz], F _s0 = 50 [Hz] and F _e0 = 7000 [Hz].

Linear prediction coefficient calculating circuit 170 receives the input vector from input terminal 10, performs a linear prediction analysis on the input vector, the linear prediction coefficients αi, i = 1, ..., determine the N _p, further the linear prediction quantized coefficients, quantized linear prediction coefficients αi ', i = 1, ... , determine the _{N p.} Here, _{N p} is the linear prediction order, for example, is 16. Further, the linear prediction coefficient calculation circuit 170 outputs the linear prediction coefficient to the weighting filter 140, and indexes corresponding to the quantized linear prediction coefficients include the linear prediction synthesis filter 130, the linear prediction inverse filter 230, and the code output circuit 290. Output to. As for quantization of the linear prediction coefficient, for example, there is a method of converting to a line spectrum pair (LSP) and quantizing. Regarding the conversion of linear prediction coefficients to LSP, “Voice Information Compression by Line Spectrum Pair (LSP) Speech Analysis and Synthesis Method” by Kashimura et al. (The IEICE Transactions A, Vol. J64-A, No. 8, pp 599-606, 1981) (Reference 3), with regard to LSP quantization, Omuro et al., “Vector quantization of LSP parameters using moving average interframe prediction” (Journal of the Institute of Electronics, Information and Communication Engineers, Vol. A, Vol. J77-A, No.3, pp.303-312, 1994) (Reference 4).

The first sound source generation circuit 110 receives the index output from the first minimization circuit 150, and stores a plurality of sound source signals (sound source vectors) as the first sound source vector corresponding to the index. Read from the table and output to the first gain circuit 160. Here, the configuration of the first sound source generation circuit 110 will be supplemented with reference to FIG. The table 1101 provided in the first sound source generation circuit 110 stores N _e sound source vectors. For example, N _e is 256. The switch 1102 inputs the index i output from the first minimizing circuit 150 via the input terminal 1103, selects the sound source vector corresponding to the index from the table, and outputs this as the first sound source vector. The signal is output to the first gain circuit 160 via the terminal 1104. In addition, for encoding a sound source signal, a method of efficiently expressing a sound source signal by a multi-pulse signal, which includes a plurality of pulses and is defined by the position of the pulse and the amplitude of the pulse, can be used. Regarding coding of excitation signals using multi-pulse signals, Ozawa et al., “MP-CELP Speech Coding Based on Multi-Pulse Vector Quantized Sound Source and Fast Search” (Journal of the Institute of Electronics, Information and Communication Engineers, pp.1655-1663) , 1996) (Reference 5). This is the end of the description of the first sound source generation circuit 110, and the description returns to FIG.

The first gain circuit 160 includes a table in which gain values are stored. The first gain circuit 160 receives the index output from the first minimizing circuit 150 and the first sound source vector output from the first sound source generation circuit 110, and receives the first gain corresponding to the index. A gain is read from the table, the first gain and the first excitation vector are multiplied, a second excitation vector is generated, and the generated second excitation vector is sent to the first bandpass filter 120. Output.

The first band pass filter 120 receives the second sound source vector output from the first gain circuit 160. The second sound source vector is band-limited to the first band by this filter to obtain a first excitation vector. The first band pass filter 120 outputs the first excitation vector to the linear prediction synthesis filter 130. Here, the first band is assumed to be F _s1 [Hz] to F _e1 [Hz]. However, F _s0 ≦ F _s1 ≦ F _e1 ≦ F _e0 . For example, F _s1 = 50 [Hz] and F _e1 = 4000 [Hz]. The first band-pass filter 120 is a high-order linear prediction filter 1 / B (z) characterized by having a band-limiting characteristic to the first band and having a linear prediction order of about 100th order. It can also be realized. Here, when N _ph is the linear prediction order and the linear prediction coefficients are β _i , i = 1,..., N _ph , the transfer function 1 / B (z) of the high-order linear prediction filter is

It is expressed. Regarding the high-order linear prediction filter, reference 2 can be referred to.

The linear prediction synthesis filter 130 includes a table in which quantized linear prediction coefficients are stored. The linear prediction synthesis filter 130 receives the first excitation vector output from the first bandpass filter 120 and the index corresponding to the quantized linear prediction coefficient output from the linear prediction coefficient calculation circuit 170. In addition, the quantized linear prediction coefficient corresponding to the index is read from the table, and the synthesis filter 1 / A (z) in which the quantized linear prediction coefficient is set is driven by the first excitation vector. A first reproduction signal (reproduction vector) is obtained. Then, the first reproduction vector is output to the first subtractor 180. Here, the transfer function 1 / A (z) of the synthesis filter is

It is expressed.

The first differentiator 180 inputs an input vector via the input terminal 10, inputs a first reproduction vector output from the linear prediction synthesis filter 130, calculates a difference between them, and calculates the first difference. The difference vector is output to the weighting filter 140 and the linear prediction inverse filter 230.

The first weighting filter 140 receives the first difference vector output from the first differentiator 180 and the linear prediction coefficient output from the linear prediction coefficient calculation circuit 170, and uses the linear prediction coefficient, A weighting filter W (z) corresponding to human auditory characteristics is generated, and the weighting filter is driven by the first difference vector, thereby obtaining a first weighted difference vector. The first weighted difference vector is output to the first minimizing circuit 150. Here, the transfer function W (z) of the weighting filter is expressed as W (z) = Q (z / γ1) / Q (z / γ2). However,

It is. γ ₁ and γ ₂ are constants, for example, γ ₁ = 0.9 and γ ₂ = 0.6. For details of the weighting filter, reference 1 can be referred to.

The first minimizing circuit 150 sequentially outputs indexes corresponding to all the first sound source vectors stored in the first sound source generating circuit 110 to the first sound source generating circuit 110 to obtain a first gain. Indexes corresponding to all the first gains stored in the circuit 160 are sequentially output to the first gain circuit 160. Further, the first weighted difference vector output from the weighting filter 140 is sequentially input, its norm is calculated, and the first sound source vector and the first gain that minimize the norm are obtained. The selected indexes are output to the code output circuit 290.

The linear prediction inverse filter 230 includes a table in which quantized linear prediction coefficients are stored. The linear prediction inverse filter 230 receives an index corresponding to the quantized linear prediction coefficient output from the linear prediction coefficient calculation circuit 170 and the first difference vector output from the first differencer 180. Further, the quantized linear prediction coefficient corresponding to the index is read from the table, and the inverse filter A (z) in which the quantized linear prediction coefficient is set is driven by the first difference vector. A residual vector of 1 is obtained. Then, the first residual vector is output to the orthogonal transformation circuit 240. Here, the transfer function A (z) of the inverse filter is

It is expressed.

The orthogonal transform circuit 240 receives the first residual vector output from the linear prediction inverse filter 230 and performs orthogonal transform on the first residual vector to obtain a second residual vector. The second residual vector is output to the band selection circuit 250. Here, as the orthogonal transform, a discrete cosine transform (DCT) can be used.

The band selection circuit 250 receives the second residual vector output from the orthogonal transform circuit 240, and uses a component included in the second band in the second residual vector, as shown in FIG. To generate N _{sbv subvectors} . An arbitrary band can be set as the second band, but here it is assumed that F _s2 [Hz] is changed to F _e2 [Hz].
However, F _s0 ≦ F _s2 ≦ F _e2 ≦ F _e0 . Here, the first band and the second band do not overlap, that is, F _e1 ≦ F _s2 . For example, F _s2 = 4000 [Hz] and F _e2 = 7000 [Hz]. The band selection circuit 250 _{outputs the} N _sbv subvectors to the orthogonal transform coefficient quantization circuit 260.

The orthogonal transform coefficient quantization circuit 260 receives N _{sbv subvectors} output from the band selection circuit 250. The orthogonal transform coefficient quantization circuit 260 includes a table storing a quantization value (shape code vector) for the shape of the subvector, a table storing a quantization value (quantization gain) for the gain of the subvector, For each of the input N _{sbv subvectors} , a shape quantization value and a gain quantization value that minimize the quantization error are selected from the table and correspond to each other. The index is output to the code output circuit 290. Here, the configuration of the orthogonal transform coefficient quantization circuit 260 will be supplemented with reference to FIG. 4, the block surrounded by the dotted line have number N _sbv, the N _sbv sub vectors in that each block is quantized. The N _{sbv subvectors}

It expresses. Since the processing for each subvector is common, the processing for e _sb , 0 (n), n = 0,..., L−1 will be described.

The subvector e _sb , 0 (n), n = 0,..., L−1 is input via the input terminal 2650. The table 2610, shape code vector ^{_{c 0 [j] (n)}} , n = 0, ..., L-1, j = 0, ..., N c, 0 -1 is stored N _c, 0 or. Here, L represents a vector length, and j represents an index. The table 2610 receives the index output from the minimizing circuit 2630 and outputs the shape code vector c ₀ ^[j] (n), n = 0,..., L−1 corresponding to the index to the gain circuit 2620. To do. The table provided in the gain circuit 2620 stores N _g , 0 quantization gains g ₀ ^[k] , k = 0,..., N _g , 0 −1. Here, k represents an index. The gain circuit 2620 inputs the shape code vector c ₀ ^[j] (n), n = 0,..., L−1 output from the table 2610, and inputs the index output from the minimization circuit 2630, The quantization gain g ₀ ^[k] corresponding to the index is read from the table, and the quantization gain g ₀ ^[k] and the shape code vector c ₀ ^[j] (n), n = 0,. Quantized subvectors e ′ _sb , 0 (n), n = 0,..., L−1 obtained by multiplying 1 are output to the differentiator 2640. The subtractor 2640 receives the sub-vector e _sb , 0 (n), n = 0,..., L−1 input via the input terminal 2650 and the quantized sub-vector e ′ _sb input from the gain circuit 2620. , 0 (n), n = 0,..., L−1, and outputs this to the minimizing circuit 2630 as a difference vector. The minimizing circuit 2630 applies all of the shape code vectors c ₀ ^[j] (n), n = 0,..., L−1, j = 0, ..., N _c , ₀ −1 stored in the table 2610. The corresponding indexes are sequentially output to the table 2610, and the indexes corresponding to all the quantization gains g ₀ ^[k] , k = 0,..., N _{g, 0} −1 stored in the gain circuit 2620 are The signals are sequentially output to the gain circuit 2620. Moreover, the difference vector are sequentially input from the differentiator 2640, the norm D ₀ calculated, the said shape code vector c ₀ norm D ₀ is the minimum ^{[j] (n), n} = 0, ..., L −1 and the quantization gain g ₀ ^[k] are selected, and indexes corresponding to these are output to the index output circuit 2660. Subvector

The same processing is performed for. The index output circuit 2660 receives the indexes output from the N _sbv minimization circuits, and outputs a set of indexes obtained by collecting them to the code output circuit 290 via the output terminal 2670. Further, the norm D ₀ is the minimum shape code vector ^{_{c 0 [j] (n)}} , n = 0, ..., L-
1 and the quantization gain g ₀ ^[k] can be determined by the following method. Norm D ₀ is

It is expressed. Where the optimal gain g ′ ₀

And norm D ₀ is

And can be transformed. Therefore, c D ₀ is minimum ^{_{0 [j] (n),}} n = 0, ..., L-1, j = 0, ..., to obtain the N _c, 0 -1, the first of (Equation 3) 2 term is the maximum ^{_{c 0 [j] (n)}} , n = 0, ..., L-1, j = 0, ..., is equivalent to finding the n _c, 0 -1. Therefore, after ^obtaining c ₀ ^[j] (n), n = 0,..., L−1, j = j _opt that maximizes the second term of (Equation 3), this c ₀ ^[j] (n ), N = 0,..., L−1, j = j _opt , g ₀ ^[k] and k = k _opt that minimize (Equation 1) are obtained. Here, as c ₀ ^[j] (n), n = 0,..., L−1, j = j _opt , a plurality of candidates are selected in descending order of the value of the second term in (Equation 3). And g ₀ ^[k] and k = k _opt that minimize (Equation 1) are obtained for each of them, and c ₀ ^[j] (n), n that minimizes the norm D ₀ among them. = 0,..., L-1, j = j _opt and g ₀ ^[k] , k = k _opt can be finally selected. Subvector

The same method can be applied to. This is the end of the description of the orthogonal transform coefficient quantization circuit 260 using FIG. 4, and the description returns to FIG.

The code output circuit 290 receives an index corresponding to the quantized linear prediction coefficient output from the linear prediction coefficient calculation circuit 170. Also, an index corresponding to each of the first excitation vector and the first gain output from the first minimizing circuit 150 is input, and N _sbv number of outputs output from the orthogonal transform coefficient quantization circuit 260 are input. Enter a set of indices consisting of shape code vectors and quantization gain indices for the subvectors. Then, as schematically shown in FIG. 29, each index is converted into a bit-sequence code and output via the output terminal 20.

The first embodiment described with reference to FIG. 1 is a case where the number of bands is 2, but a case where the number of bands is expanded to 3 or more will be described below.

FIG. 1 can be rewritten as shown in FIG. Here, the first encoding circuit 1001 in FIG. 5 is equivalent to FIG. 6, and the second encoding circuit 1002 in FIG. 5 is equivalent to FIG. The blocks are the same as the blocks described in FIG.

The second embodiment of the present invention is realized by expanding the number of bands to 3 in the first embodiment. The configuration of the speech and music signal encoding apparatus according to the second embodiment of the present invention can be represented by the block diagram shown in FIG. Here, the first encoding circuit 1001 is equivalent to FIG. 6, the second encoding circuit 1002 is equivalent to FIG. 6, and the third encoding circuit 1003 is equivalent to FIG. The code output circuit 2901 receives the index output from the linear prediction coefficient calculation circuit 170, receives the index output from the first encoding circuit 1001, and receives the index output from the second encoding circuit 1002. The set of indexes output from the third encoding circuit 1003 is input. Each index is converted into a bit-sequence code and output via the output terminal 20.

The third embodiment of the present invention is realized by extending the number of bands to N in the first embodiment. The configuration of the speech and music signal encoding apparatus according to the third embodiment of the present invention can be represented by the block diagram shown in FIG. Here, the first encoding circuit 1001 to the (N-1) th encoding circuit 1004 are equivalent to FIG. 6, and the Nth encoding circuit 1005 is equivalent to FIG. The code output circuit 2902 receives an index output from the linear prediction coefficient calculation circuit 170, inputs an index output from each of the first to N-1th encoding circuits 1004 to 1001, and An index set output from the N encoding circuits 1005 is input. Each index is converted into a bit-sequence code and output via the output terminal 20.

In the first embodiment, the first encoding circuit 1001 in FIG. 5 is based on an encoding method using an AbS (Analysis-by-Synthesis) method. On the other hand, an encoding method other than the Abs method can be applied. Hereinafter, a case where an encoding method using time-frequency conversion is applied to the first encoding circuit 1001 as an encoding method other than the Abs method will be described.

The fourth embodiment of the present invention is realized by applying an encoding method using time-frequency conversion in the first embodiment. The configuration of the speech and music signal encoding apparatus according to the fourth embodiment of the present invention can be represented by the block diagram shown in FIG. Here, the first encoding circuit 1011 is equivalent to FIG. 10, and the second encoding circuit 1002 is equivalent to FIG. Among the blocks constituting FIG. 10, the linear prediction inverse filter 230, the orthogonal transform circuit 240, the band selection circuit 250, and the orthogonal transform coefficient quantization circuit 260 are the same as the blocks described in FIG. The orthogonal transform coefficient inverse quantization circuit 460, the orthogonal inverse transform circuit 440, and the linear prediction synthesis filter 131 are blocks constituting a speech and music decoding device corresponding to the first embodiment according to a ninth embodiment described later. The same. The description of the orthogonal transform coefficient inverse quantization circuit 460, the orthogonal inverse transform circuit 440, and the linear prediction synthesis filter 131 will be omitted because it will be described in the description of the ninth embodiment using FIG. The code output circuit 2903 receives the index output from the linear prediction coefficient calculation circuit 170, receives the index set output from the first encoding circuit 1011, and outputs from the second encoding circuit 1002. Enter a set of indexes. Each index is converted into a bit-sequence code and output via the output terminal 20.

The fifth embodiment of the present invention is realized by expanding the number of bands to 3 in the fourth embodiment. The configuration of the speech and music signal encoding apparatus according to the fifth embodiment of the present invention can be represented by the block diagram shown in FIG. Here, the first encoding circuit 1011 is equivalent to FIG. 10, the second encoding circuit 1012 is equivalent to FIG. 10, and the third encoding circuit 1003 is equivalent to FIG. The code output circuit 2904 receives the index output from the linear prediction coefficient calculation circuit 170, receives the index set output from the first encoding circuit 1011, and outputs the index set from the second encoding circuit 1012. An index set is input, and an index set output from the third encoding circuit 1003 is input. Each index is converted into a bit-sequence code and output via the output terminal 20.

The sixth embodiment of the present invention is realized by expanding the number of bands to N in the fourth embodiment. The configuration of the speech and music signal encoding apparatus according to the sixth embodiment of the present invention can be represented by the block diagram shown in FIG. Here, each of the first to N-1th encoding circuits 1011 to 1014 is equivalent to FIG. 10, and the Nth encoding circuit 1005 is equivalent to FIG. The code output circuit 2905 receives the index output from the linear prediction coefficient calculation circuit 170, and receives the set of indexes output from each of the N-1th encoding circuit 1014 from the first encoding circuit 1011. The index set output from the Nth encoding circuit 1005 is input. Each index is converted into a bit-sequence code and output via the output terminal 20.

FIG. 14 is a block diagram showing the configuration of a speech and music signal encoding apparatus according to the seventh embodiment of the present invention. A block surrounded by a dotted line in the figure is called a pitch prediction filter, and FIG. 14 is obtained by adding the pitch prediction filter to FIG. Hereinafter, the memory circuit 510, the pitch signal generation circuit 112, the third gain circuit 162, the adder 184, the first minimization circuit 550, and the code output circuit 590, which are different blocks from those in FIG. 1, will be described.

The storage circuit 510 receives the fifth sound source signal from the adder 184 and holds it. The storage circuit 510 outputs the fifth sound source signal input and held in the past to the pitch signal generation circuit 112.

The pitch signal generation circuit 112 receives the past fifth sound source signal held in the storage circuit 510 and the index output from the first minimization circuit 550. The index specifies a delay d. Then, as shown in FIG. 30, in the past fifth sound source signal, a signal for L samples corresponding to the vector length is cut out from a point d samples past from the start point of the current frame, and the first pitch vector is obtained. Generate. Here, when d <L, a signal for d samples is cut out, and the cut out d samples are repeatedly connected to generate a first pitch vector having a vector length of L samples. The pitch signal generation circuit 112 outputs the first pitch vector to the third gain circuit 162.

The third gain circuit 162 includes a table in which gain values are stored. The third gain circuit 162 receives the index output from the first minimization circuit 550 and the first pitch vector output from the pitch signal generation circuit 112, and obtains a third gain corresponding to the index. The second pitch vector is read from the table, multiplied by the third gain and the first pitch vector to generate a second pitch vector, and the generated second pitch vector is output to the adder 184.

The adder 184 inputs the second sound source vector output from the first gain circuit 160 and the second pitch vector output from the third gain circuit 162, calculates the sum of these, The fifth sound source vector is output to the first band pass filter 120.

The first minimizing circuit 550 sequentially outputs indexes corresponding to all the first sound source vectors stored in the first sound source generating circuit 110 to the first sound source generating circuit 110, and the pitch signal generating circuit. 112, indexes corresponding to all delays d within the range defined in 112 are sequentially output to the pitch signal generation circuit 112, and indexes corresponding to all the first gains stored in the first gain circuit 160 are The first gain circuit 160 is sequentially output, and indexes corresponding to all the third gains stored in the third gain circuit 162 are sequentially output to the third gain circuit 162. In addition, the first weighted difference vector output from the weighting filter 140 is sequentially input, the norm thereof is calculated, and the first sound source vector, the delay d, the first, such that the norm is minimized. A gain of 1 and the third gain are selected, and indexes corresponding to these are collectively output to the code output circuit 590.

The code output circuit 590 inputs an index corresponding to the quantized linear prediction coefficient output from the linear prediction coefficient calculation circuit 170. Also, the index corresponding to each of the first sound source vector, the delay d, the first gain, and the third gain output from the first minimizing circuit 550 is input, and the orthogonal transform coefficient quantizing circuit 260 outputs the index. A set of indexes composed of shape code vectors and quantization gain indexes for the N _{sbv subvectors} to be output are input. Each index is converted into a bit-sequence code and output via the output terminal 20.

FIG. 15 is a block diagram showing the configuration of a speech and music signal encoding apparatus according to the eighth embodiment of the present invention. In the following, the downsampling circuit 780, the first linear prediction coefficient calculation circuit 770, the first linear prediction synthesis filter 132, the third differentiator 183, the upsampling circuit 781, the first block, which are different blocks from FIG. Differentiator 180,
The second linear prediction coefficient calculation circuit 771, the third linear prediction coefficient calculation circuit 772, the linear prediction inverse filter 730, and the code output circuit 790 will be described.

The down-sampling circuit 780 inputs an input vector from the input terminal 10 and obtains a second input vector having a first band obtained by down-sampling the input vector from the first linear prediction coefficient calculation circuit 770 and the third Output to the differentiator 183. Here, the first band is set to F _s1 [Hz] to F _e1 [Hz] as in the first embodiment, and the input vector band is set to F _s0 [Hz] to F _e0 [Hz] (third Band). Regarding the configuration of the downsampling circuit, section 4.1.1 of the document entitled “Multirate Systems and Filter Banks” by PP Vaidyanathan (Reference 6) can be referred to.

The first linear prediction coefficient calculation circuit 770 receives the second input vector from the downsampling circuit 780, performs linear prediction analysis on the second input vector, and performs the first linear having the first band. A prediction coefficient is obtained, and further, the first linear prediction coefficient is quantized to obtain a first quantized linear prediction coefficient. The first linear prediction coefficient calculation circuit 770 outputs the first linear prediction coefficient to the first weighting filter 140, and an index corresponding to the first quantized linear prediction coefficient is output to the first linear prediction synthesis filter. 132, the linear prediction inverse filter 730, the third linear prediction coefficient calculation circuit 772, and the code output circuit 790.

The first linear prediction synthesis filter 132 includes a table in which the first quantized linear prediction coefficients are stored. The first linear prediction synthesis filter 132 uses the fifth excitation vector output from the adder 184 and the index corresponding to the first quantized linear prediction coefficient output from the first linear prediction coefficient calculation circuit 770. input. Further, the first quantized linear prediction coefficient corresponding to the index is read from the table, and the synthesis filter in which the first quantized linear prediction coefficient is set is driven by the fifth excitation vector. Then, a first reproduction vector having a first band is obtained. Then, the first reproduction vector is output to the third differentiator 183 and the upsampling circuit 781.

The third differentiator 183 inputs the first reproduction vector output from the first linear prediction synthesis filter 132 and the second input vector output from the downsampling circuit 780, and calculates the difference between them. This is output to the weighting filter 140 as the second difference vector.

The upsampling circuit 781 receives the first reproduction vector output from the first linear prediction synthesis filter 132 and upsamples it to obtain a third reproduction vector having a third band. Here, the third band is from F _s0 [Hz] to F _e0 [Hz]. The upsample circuit 781 outputs the third reproduction vector to the first subtractor 180. For the configuration of the upsample circuit, refer to section 4.1.1 of the document (Reference 6) entitled “Multirate Systems and Filter Banks” by PP Vaidyanathan.

The first differentiator 180 receives an input vector via the input terminal 10, inputs a third reproduction vector output from the upsampling circuit 781, calculates a difference between them, and uses this as a first difference. The vector is output to the linear prediction inverse filter 730 as a vector.

The second linear prediction coefficient calculation circuit 771 receives an input vector from the input terminal 10, performs linear prediction analysis on the input vector, obtains a second linear prediction coefficient having a third band, and The second linear prediction coefficient is output to the third linear prediction coefficient calculation circuit 772.

The third linear prediction coefficient calculation circuit 772 includes a table in which the first quantized linear prediction coefficient is stored. The third linear prediction coefficient calculation circuit 772 includes the second linear prediction coefficient output from the second linear prediction coefficient calculation circuit 771 and the first quantization output from the first linear prediction coefficient calculation circuit 770. Input an index corresponding to the linear prediction coefficient. Then, the first quantized linear prediction coefficient corresponding to the index is read from the table, the first quantized linear prediction coefficient is converted to LSP, and further, the sampling frequency is converted, thereby converting the input signal. A first LSP corresponding to the sampling frequency is obtained. Further, the second linear prediction coefficient is converted into an LSP to obtain a second LSP. A difference between the second LSP and the first LSP is calculated, and this is defined as a third LSP. Here, Japanese Patent Application No. 9-202475 (Document 7) can be referred to for sampling frequency conversion of LSP. The third LSP is quantized and converted into a linear prediction coefficient to obtain a third quantized linear prediction coefficient having a third band. The index corresponding to the third quantized linear prediction coefficient is output to the linear prediction inverse filter 730 and the code output circuit 790.

The linear prediction inverse filter 730 includes a first table storing a first quantized linear prediction coefficient and a second table storing a third quantized linear prediction coefficient. The linear prediction inverse filter 730 includes a first index corresponding to the first quantized linear prediction coefficient output from the first linear prediction coefficient calculation circuit 770 and a first index output from the third linear prediction coefficient calculation circuit 772. The second index corresponding to the quantized linear prediction coefficient of 3 and the first difference vector output from the first differentiator 180 are input. The linear prediction inverse filter 730 reads the first quantized linear prediction coefficient corresponding to the first index from the first table, converts the coefficient into an LSP, and further converts the input signal to the sampling frequency. The first LSP corresponding to the sampling frequency is obtained. Then, the third quantized linear prediction coefficient corresponding to the second index is read from the second table and converted into an LSP to obtain a third LSP. Next, the first LSP and the third LSP are added to obtain a second LSP. The linear prediction inverse filter 730 converts the second LSP into a linear prediction coefficient, obtains a second quantized linear prediction coefficient, and performs an inverse filter in which the second quantized linear prediction coefficient is set, By driving with a difference vector of 1, a first residual vector is obtained. Then, the first residual vector is output to the orthogonal transformation circuit 240.

The code output circuit 790 inputs an index corresponding to the first quantized linear prediction coefficient output from the first linear prediction coefficient calculation circuit 770 and outputs a third output from the third linear prediction coefficient calculation circuit 772. An index corresponding to each of the first excitation vector, the delay d, the first gain, and the third gain output from the first minimizing circuit 550 is input. , And a set of indices composed of shape code vectors and quantization gain indices for N _{sbv subvectors} output from the orthogonal transform coefficient quantization circuit 260. Each index is converted into a bit-sequence code and output via the output terminal 20.

FIG. 16 is a block diagram showing a configuration of a speech and music signal decoding apparatus corresponding to the first embodiment according to the ninth embodiment of the present invention. The decoding apparatus inputs a bit sequence code from an input terminal 30.

The code input circuit 410 converts the bit sequence code input from the input terminal 30 into an index. The index corresponding to the first sound source vector is output to the first sound source generation circuit 110. The index corresponding to the first gain is output to the first gain circuit 160. The index corresponding to the quantized linear prediction coefficient is output to the linear prediction synthesis filter 130 and the linear prediction synthesis filter 131. A set of indexes obtained by collecting the index N _sbv corresponding to each of the shape code vector and the quantization gain for the _subvector for the number of _subvectors is output to the orthogonal transform coefficient inverse quantization circuit 460.

The first sound generator 110 receives an index output from the code input circuit 410, reads a first sound vector corresponding to the index from a table storing a plurality of sound vectors, Output to the gain circuit 160.

The first gain circuit 160 includes a table storing quantization gains. The first gain circuit 160 receives the index output from the code input circuit 410 and the first sound source vector output from the first sound source generation circuit 110, and sets the first gain corresponding to the index to the first gain circuit 160. It reads from the table, multiplies the first gain and the first sound source vector to generate a second sound source vector, and outputs the generated second sound source vector to the first band pass filter 120.

The first band pass filter 120 receives the second sound source vector output from the first gain circuit 160. The second sound source vector is band-limited to the first band by this filter to obtain a first excitation vector. The first band pass filter 120 outputs the first excitation vector to the linear prediction synthesis filter 130.

The configuration of the orthogonal transform coefficient inverse quantization circuit 460 will be described with reference to FIG. In FIG. 18, there are N _sbv blocks surrounded by a dotted line. N _sbv quantization _subvectors defined in the band selection circuit 250 of FIG. 1 in each block

Is decrypted. Since the decoding process for each quantization subvector is common, the process for e ′ _sb , 0 (n), n = 0,..., L−1 will be described. Quantization subvector _{e 'sb, 0 (n)} , n = 0, ..., L-1 , as well as the processing of the orthogonal transform coefficient quantizing circuit 260 in FIG. 1, the shape code vector c ₀ ^[j] ( n), n = 0,..., L−1 and the quantization gain g ₀ ^[k] . Here, j and k represent indexes. The index input circuit 4630 receives, via an input terminal 4650, an index set _if composed of shape code vectors and quantization gain indices for the N _sbv quantization _subvectors output from the code input circuit 410. To do. The index that specifies the from the set i _f of the index, the shape code vector ^{_{c 0 [j] (n)}} , n = 0, ..., index i _sbs, 0 and quantization gain g0 specifying the L-1 [k] i _sbg , 0 is taken out, i _sb s, 0 is output to the table 4610, and i _sbg , 0 is output to the gain circuit 4620. The table ^{_{4610, c 0 [j] (}} n), n = 0, ..., L-1, j = 0, ..., N c, 0 -1 is stored. The table 4610 receives the index i _sbs , 0 output from the index input circuit 4630, and the shape code vector c ₀ ^[j] (n), n = 0,..., L−1, j corresponding to i _sbs , 0. = I _sbs , 0 is output to the gain circuit 4620. The table provided in the gain circuit 4620 stores g ₀ ^[k] , k = 0,..., N _{g, 0} −1. The gain circuit 4620 receives c ₀ ^[j] (n), n = 0,..., L−1, j = i _sbs , 0 output from the table 4610, and the index i output from the index input circuit 4630. _sbg, 0 type a, i _sbg, 0 corresponding to the quantized gain ^{_{g 0 [k], k =}} i sbg, 0 read from the ^{_{table, c 0 [j] (n}} ), n = 0, ..., Quantized subvector e ′ _sb , 0 (n), n = 0,..., L obtained by multiplying L−1, j = i _sbg , 0 and g ₀ ^[k] , k = i _sbg , 0 −1 is output to the full-band vector generation circuit 4640. The full-band vector generation circuit 4640 receives the quantized subvector e ′ _sb , 0 (n), n = 0,..., L−1 output from the gain circuit 4620. The all-band vector generation circuit 4640 is obtained by the same processing as e ′ _sb , 0 (n), n = 0,..., L−1.

Enter. And as shown in FIG.
The N _sbv quantization subvectors

1 is arranged in the second band defined by the band selection circuit 250 in FIG. 1, and a zero vector is arranged in addition to the second band, so that the whole band (for example, the sampling frequency of the reproduction signal is 16 kHz). 2), a second excitation vector corresponding to the 8 kHz band) is generated and output to the orthogonal inverse transform circuit 440 via the output terminal 4660.

The orthogonal inverse transform circuit 440 receives the second excitation vector output from the orthogonal transform coefficient inverse quantization circuit 460 and orthogonally inverse transforms the second excitation vector to obtain a third excitation vector. Then, the third excitation vector is output to the linear prediction synthesis filter 131. Here, as the inverse orthogonal transform, an inverse discrete cosine transform (IDCT) can be used.

The linear prediction synthesis filter 130 includes a table in which quantized linear prediction coefficients are stored. The linear prediction synthesis filter 130 receives the first excitation vector output from the first bandpass filter 120 and the index corresponding to the quantized linear prediction coefficient output from the code input circuit 410. In addition, the quantized linear prediction coefficient corresponding to the index is read from the table, and the synthesis filter 1 / A (z) in which the quantized linear prediction coefficient is set is driven by the first excitation vector. To obtain a first reproduction vector. Then, the first reproduction vector is output to the adder 182.

The linear prediction synthesis filter 131 includes a table in which quantized linear prediction coefficients are stored. The linear prediction synthesis filter 131 receives the third excitation vector output from the orthogonal inverse transform circuit 440 and the index corresponding to the quantized linear prediction coefficient output from the code input circuit 410. In addition, the quantized linear prediction coefficient corresponding to the index is read from the table, and the synthesis filter 1 / A (z) in which the quantized linear prediction coefficient is set is driven by the third excitation vector. Then, a second reproduction vector is obtained. Then, the second reproduction vector is output to the adder 182.

The adder 182 inputs the first reproduction vector output from the linear prediction synthesis filter 130 and the second reproduction vector output from the linear prediction synthesis filter 131, calculates the sum of these, 3 as the reproduction vector 3 through the output terminal 40.

The ninth embodiment described with reference to FIG. 16 is a case where the number of bands is 2, but a case where the number of bands is expanded to 3 or more will be described below.

FIG. 16 can be rewritten as shown in FIG. Here, the first decoding circuit 1051 in FIG. 19 is equivalent to FIG. 20, the second decoding circuit 1052 in FIG. 19 is equivalent to FIG. 21, and each block constituting FIG. 20 and FIG. This is the same as each block described in FIG.

The tenth embodiment of the present invention is realized by extending the number of bands to 3 in the ninth embodiment. The configuration of the speech and music signal decoding apparatus according to the tenth embodiment of the present invention can be represented by the block diagram shown in FIG. Here, the first decoding circuit 1051 is equivalent to FIG. 20, the second decoding circuit 1052 is equivalent to FIG. 20, and the third decoding circuit 1053 is equivalent to FIG. The code input circuit 4101 converts the code of the bit sequence input from the input terminal 30 into an index, and converts the index corresponding to the quantized linear prediction coefficient to the first decoding circuit 1051, the second decoding circuit 1052, and the third decoding. An output corresponding to the sound source vector and the gain is output to the first decoding circuit 1051 and the second decoding circuit 1052, and a set of the index corresponding to the shape code vector and the quantization gain for the subvector is output to the circuit 1053. 3 to the decoding circuit 1053.

The eleventh embodiment of the present invention is realized by expanding the number of bands to N in the ninth embodiment. The configuration of the speech and music signal decoding apparatus according to the eleventh embodiment of the present invention can be represented by the block diagram shown in FIG. Here, each of the first decoding circuit 1051 to the (N-1) th decoding circuit 1054 is equivalent to FIG. 20, and the Nth decoding circuit 1055 is equivalent to FIG. The code input circuit 4102 converts the code of the bit sequence input from the input terminal 30 into an index, and converts the index corresponding to the quantized linear prediction coefficient from the first decoding circuit 1051 to the (N−1) -th decoding circuit 1054 and the N-th decoding circuit. Output to each of the decoding circuits 1055, and output the sound source vector and the index corresponding to the gain from each of the first decoding circuit 1051 to each of the (N-1) th decoding circuits 1054, and convert the shape code vector and quantization gain for the subvector. The corresponding set of indexes is output to the Nth decoding circuit 1055.

In the ninth embodiment, the first decoding circuit 1051 in FIG. 19 is based on a decoding method corresponding to an encoding method using the Abs method, but for the first decoding circuit 1051, A decoding method corresponding to an encoding method other than the Abs method can also be applied. Below, the case where the decoding system corresponding to the encoding system using time frequency conversion is applied with respect to the 1st decoding circuit 1051 is demonstrated.

The twelfth embodiment of the present invention is realized by applying a decoding scheme corresponding to the encoding scheme using time-frequency conversion in the ninth embodiment. The configuration of the speech and music signal decoding apparatus according to the twelfth embodiment of the present invention can be represented by the block diagram shown in FIG. Here, the first decoding circuit 1061 is equivalent to FIG. 21, and the second decoding circuit 1052 is equivalent to FIG. The code input circuit 4103 converts the code of the bit sequence input from the input terminal 30 into an index, outputs the index corresponding to the quantized linear prediction coefficient to the first decoding circuit 1061 and the second decoding circuit 1052, and A set of indexes corresponding to the shape code vector and the quantization gain for the vector is output to first decoding circuit 1061 and second decoding circuit 1052.

The thirteenth embodiment of the present invention is realized by expanding the number of bands to three in the twelfth embodiment. The configuration of the speech and music signal decoding apparatus according to the thirteenth embodiment of the present invention can be represented by the block diagram shown in FIG. Here, the first decoding circuit 1061 is equivalent to FIG. 21, the second decoding circuit 1062 is equivalent to FIG. 21, and the third decoding circuit 1053 is equivalent to FIG. The code input circuit 4104 converts the code of the bit sequence input from the input terminal 30 into an index, and converts the index corresponding to the quantized linear prediction coefficient to the first decoding circuit 1061, the second decoding circuit 1062, and the third decoding. The output to the circuit 1053 is output to the first decoding circuit 1061, the second decoding circuit 1062, and the third decoding circuit 1053, the set of indices corresponding to the shape code vector and the quantization gain for the subvector.

The fourteenth embodiment of the present invention is realized by extending the number of bands to N in the twelfth embodiment. The configuration of the speech and music signal decoding apparatus according to the fourteenth embodiment of the present invention can be represented by the block diagram shown in FIG. Here, each of the first decoding circuit 1061 to the (N-1) th decoding circuit 1064 is equivalent to FIG. 21, and the Nth decoding circuit 1055 is equivalent to FIG. The code input circuit 4105 converts the code of the bit sequence input from the input terminal 30 into an index, and converts the index corresponding to the quantized linear prediction coefficient from the first decoding circuit 1061 to the (N−1) -th decoding circuit 1064 and the N-th decoding circuit. Are output to each of the first decoding circuit 1055, and a set of indexes corresponding to the shape code vector and the quantization gain for the sub-vector is obtained from the first decoding circuit 1061 to the (N−1) th decoding circuit 1064 and the Nth decoding circuit 1055. Output to each.

FIG. 27 is a block diagram showing a configuration of a speech and music signal decoding apparatus corresponding to the seventh embodiment according to the fifteenth embodiment of the present invention. 27, the blocks different from the ninth embodiment of FIG. 16 are a storage circuit 510, a pitch signal generation circuit 112, a third gain circuit 162, an adder 184, and a sign input circuit 610. Since the pitch signal generation circuit 112, the third gain circuit 162, and the adder 184 are the same as those in FIG. 14, the description thereof will be omitted, and the code input circuit 610 will be described.

The code input circuit 610 converts the code of the bit sequence input from the input terminal 30 into an index. The index corresponding to the first sound source vector is output to the first sound source generation circuit 110. The index corresponding to the delay d is output to the pitch signal generation circuit 112. The index corresponding to the first gain is output to the first gain circuit 160. The index corresponding to the third gain is output to the third gain circuit 162. The index corresponding to the quantized linear prediction coefficient is output to the linear prediction synthesis filter 130 and the linear prediction synthesis filter 131. A set of indexes obtained by collecting indexes corresponding to each of the shape code vector and the quantization gain for the _subvector for N _{sbv subvectors} is output to orthogonal transform coefficient inverse quantization circuit 460.

FIG. 28 is a block diagram showing a configuration of a speech and music signal decoding apparatus corresponding to the eighth embodiment according to the sixteenth embodiment of the present invention. Hereinafter, the code input circuit 810, the first linear prediction coefficient synthesis filter 132, the upsampling circuit 781, and the second linear prediction synthesis filter 831 which are different blocks from those in FIG. 27 will be described.

The code input circuit 810 converts the code of the bit sequence input from the input terminal 30 into an index. The index corresponding to the first sound source vector is output to the first sound source generation circuit 110. The index corresponding to the delay d is output to the pitch signal generation circuit 112. The index corresponding to the first gain is output to the first gain circuit 160. The index corresponding to the third gain is output to the third gain circuit 162. The index corresponding to the first quantized linear prediction coefficient is output to the first linear prediction synthesis filter 132 and the second linear prediction synthesis filter 831. The index corresponding to the third quantized linear prediction coefficient is output to the second linear prediction synthesis filter 831. A set of indexes obtained by collecting indexes corresponding to each of the shape code vector and the quantization gain for the _subvector for N _{sbv subvectors} is output to orthogonal transform coefficient inverse quantization circuit 460.

The first linear prediction synthesis filter 132 includes a table in which the first quantized linear prediction coefficients are stored. The first linear prediction synthesis filter 132 receives the fifth excitation vector output from the adder 184 and the index corresponding to the first quantized linear prediction coefficient output from the code input circuit 810. Further, the first quantized linear prediction coefficient corresponding to the index is read from the table, and the synthesis filter in which the first quantized linear prediction coefficient is set is driven by the fifth excitation vector. Then, a first reproduction vector having a first band is obtained. Then, the first reproduction vector is output to the upsampling circuit 781.

The upsampling circuit 781 receives the first reproduction vector output from the first linear prediction synthesis filter 132 and upsamples it to obtain a third reproduction vector having a third band. The third reproduction vector is output to the first adder 182.

The second linear prediction synthesis filter 831 stores a first table storing a first quantized linear prediction coefficient having a first band and a third quantized linear prediction coefficient having a third band. The second table is provided. The second linear prediction synthesis filter 831 includes a third excitation vector output from the orthogonal inverse transform circuit 440 and a first index corresponding to the first quantized linear prediction coefficient output from the code input circuit 810. , And the second index corresponding to the third quantized linear prediction coefficient. The second linear prediction synthesis filter 831 reads the first quantized linear prediction coefficient corresponding to the first index from the first table, converts it to LSP, and further converts this to the sampling frequency. Thus, the first LSP corresponding to the sampling frequency of the third reproduction vector is obtained. Next, the third quantized linear prediction coefficient corresponding to the second index is read from the second table and converted into an LSP to obtain a third LSP. Then, the second LSP obtained by adding the first LSP and the third LSP is converted into a linear prediction coefficient to obtain a second linear prediction coefficient. The second linear prediction synthesis filter 831 obtains a second reproduction vector having a third band by driving the synthesis filter in which the second linear prediction coefficient is set with the third excitation vector. . Then, the second reproduction vector is output to the adder 182.

The adder 182 inputs the third reproduction vector output from the upsampling circuit 781 and the second reproduction vector output from the second linear prediction synthesis filter 831, calculates the sum of these, The fourth reproduction vector is output via the output terminal 40.

It is a block diagram which shows the structure of the audio | voice music signal encoding apparatus by 1st Example of this invention. 2 is a block diagram showing a configuration of a first sound source generation circuit 110. FIG. 6 is a diagram for explaining a method of generating a subvector in band selection circuit 250. FIG. 3 is a block diagram showing a configuration of an orthogonal transform coefficient quantization circuit 260. FIG. It is a block diagram equivalent to FIG. 1 which shows the structure of the audio | voice music signal encoding apparatus by 1st Example of this invention. FIG. 6 is a block diagram showing a configuration of a first encoding circuit 1001 in FIG. 5. FIG. 6 is a block diagram showing a configuration of a second encoding circuit 1002 in FIG. 5. It is a block diagram which shows the structure of the audio | voice music signal encoding apparatus by the 2nd Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal encoding apparatus by the 3rd Example of this invention. FIG. 12 is a block diagram showing a configuration of a first encoding circuit 1011 in FIG. 11. It is a block diagram which shows the structure of the audio | voice music signal encoding apparatus by the 4th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal encoding apparatus by the 5th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal encoding apparatus by the 6th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal encoding apparatus by the 7th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal encoding apparatus by the 8th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal decoding apparatus by the 9th Example of this invention. It is a figure for demonstrating the method to produce | generate a 2nd excitation vector in the orthogonal transformation coefficient inverse quantization circuit 460. FIG. 4 is a block diagram showing a configuration of an orthogonal transform coefficient inverse quantization circuit 460. FIG. It is a block diagram equivalent to FIG. 16 which shows the structure of the audio | voice music signal decoding apparatus by the 9th Example of this invention. FIG. 20 is a block diagram illustrating a configuration of a first decoding circuit 1051 in FIG. 19. FIG. 20 is a block diagram illustrating a configuration of a second decoding circuit 1052 in FIG. 19. It is a block diagram which shows the structure of the audio | voice music signal decoding apparatus by the 10th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal decoding apparatus by 11th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal decoding apparatus by the 12th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal decoding apparatus by the 13th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal decoding apparatus by the 14th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal decoding apparatus by 15th Example of this invention. It is a block diagram which shows the structure of the audio | voice music signal decoding apparatus by the 16th Example of this invention. FIG. 10 is a diagram for explaining a correspondence between an index and a bit sequence code in a code output circuit 290; 6 is a diagram for explaining a method of generating a first pitch vector in the pitch signal generation circuit 112. FIG. It is a block diagram which shows embodiment of the audio | voice music signal encoding apparatus by a conventional method. It is a block diagram which shows embodiment of the audio | voice music signal decoding apparatus by a conventional method.

Explanation of symbols

10, 30 Input terminal 20, 40 Output terminal 110 First sound generator circuit 111 Second sound generator circuit 160 First gain circuit 161 Second gain circuit 120 First band pass filter 121 Second band pass filter 182, 184 Adder 180 First differencer 181 Second differencer 183 Third differencer 170 Linear prediction coefficient calculation circuit 770 First linear prediction coefficient calculation circuit 771 Second linear prediction coefficient calculation circuit 772 Third Linear prediction coefficient calculation circuit 130 linear prediction synthesis filter 131 linear prediction synthesis filter 132 first linear prediction synthesis filter 831 second linear prediction synthesis filter 140 weighting filter 141 weighting filters 150 and 550 first minimizing circuit 151 Second minimizing circuits 230 and 730 linear prediction inverse filter 240 orthogonal transform circuit 250 Area selection circuit 260 Orthogonal transform coefficient quantization circuit 440 Orthogonal inverse transform circuit 460 Orthogonal transform coefficient inverse quantization circuit 190, 290, 590, 790 Code output circuit 310, 410, 610, 810 Code input circuit 780 Downsample circuit 781 Upsample Circuit 510 Memory circuit 112 Pitch signal generation circuit 162 Third gain circuit 1101 Table 1102 Switch 1103 Input terminal 1104 Output terminal 2650, 2651 Input terminal 2610, 2611 Table 2620, 2621 Gain circuit 2630, 2631 Minimization circuit 2640, 2641 Differentiator 2660 Index output circuit 2670 Output terminals 1001 and 1011 First encoding circuit 1002 and 1012 Second encoding circuit 1003 Third encoding circuit 1004 and 1014 N-1th encoding time 1005 Nth encoding circuit 2901, 2902, 2903, 2904, 2905 Code output circuit 1801, 1802 Differentiator 4610, 4611 Table 4620, 4621 Gain circuit 4630 Index input circuit 4640 Full-band vector generation circuit 4650 Input terminal 4660 Output terminal 1051 , 1061 First decoding circuit 1052, 1062 Second decoding circuit 1053 Third decoding circuit 1054, 1064 N-1 decoding circuit 1055 Nth decoding circuit 4101, 4102, 4103, 4104, 4105 Code input circuit 1821 , 1822 Adder

Claims (6)

Receiving a first input signal sampled at a first sampling frequency;
Down-sampling the first input signal to a second sampling frequency to generate a second input signal;
Obtaining a first linear prediction coefficient of a second sampling frequency from the second input signal;
A first excitation signal having the second sampling frequency for generating a first reproduction signal corresponding to the second input signal by driving a synthesis filter in which the first linear prediction coefficient is set is obtained. ,
Generating a second reproduction signal by up-sampling the first reproduction signal to the first sampling frequency;
A frequency domain parameter corresponding to a linear prediction coefficient obtained from the first input signal and a frequency corresponding to a second linear prediction coefficient obtained by converting the first linear prediction coefficient into a first sampling frequency by sampling frequency conversion. A third linear prediction coefficient is obtained from the difference from the region parameter,
Obtaining a fourth linear prediction coefficient from a sum of a frequency domain parameter corresponding to the second linear prediction coefficient and a frequency domain parameter corresponding to the third linear prediction coefficient;
A residual signal is generated by driving an inverse filter in which the fourth linear prediction coefficient is set by a difference signal between the first input signal and the second reproduction signal;
Orthogonally transform the residual signal;
A speech music signal characterized by outputting a component corresponding to at least a part of a band of the orthogonally transformed residual signal, the first linear prediction coefficient, the third linear prediction coefficient, and the excitation signal. Encoding device. A first linear prediction coefficient of a second sampling frequency obtained from a second input signal generated by down-sampling the first input signal sampled at a first sampling frequency to a second sampling frequency; The first sampling frequency of the second sampling frequency obtained for driving the synthesis filter in which the first linear prediction coefficient is set to generate the first reproduction signal corresponding to the second input signal. A second linear prediction obtained by sampling frequency conversion of the excitation signal, the frequency domain parameter corresponding to the linear prediction coefficient obtained from the first input signal, and the first linear prediction coefficient into a first sampling frequency A third linear prediction coefficient obtained from the difference from the frequency domain parameter corresponding to the coefficient, and the first reproduction signal as the first sampling frequency. A frequency domain parameter corresponding to the second linear prediction coefficient and the third linear prediction coefficient based on a difference signal between the second reproduction signal generated by up-sampling to a number and the first input signal. Of the residual signal after orthogonal transformation obtained by orthogonal transformation of the residual signal generated by driving the inverse filter in which the fourth linear prediction coefficient obtained from the sum with the corresponding frequency domain parameter is set Receiving a signal including a component corresponding to at least a part of the band;
A first reproduction signal is generated by up-sampling a signal obtained by driving a first linear prediction synthesis filter with the first excitation signal to the first sampling frequency;
Generating a second excitation signal having a first sampling frequency by performing orthogonal inverse transformation on a component corresponding to at least a part of the band of the residual signal after orthogonal transformation;
A second linear prediction synthesis filter in which a signal obtained by adding a frequency domain parameter corresponding to the first linear prediction coefficient and a frequency domain parameter corresponding to the third linear prediction coefficient is set; A second reproduction signal is generated by driving with the excitation signal of
A speech and music signal decoding apparatus, wherein a speech and music signal is generated by adding the first reproduction signal and the second reproduction signal. The component corresponding to at least a part of the band of the orthogonally transformed residual signal, the first linear prediction coefficient of the second sampling frequency, the third linear prediction coefficient of the first sampling frequency, and the second sampling frequency Receiving a signal including a first excitation signal;
A first reproduction signal is generated by up-sampling a signal obtained by driving a first linear prediction synthesis filter with the first excitation signal to the first sampling frequency;
Generating a second excitation signal having a first sampling frequency by performing orthogonal inverse transformation on a component corresponding to at least a part of the band of the residual signal after orthogonal transformation;
A second linear prediction synthesis filter in which a signal obtained by adding a frequency domain parameter corresponding to the first linear prediction coefficient and a frequency domain parameter corresponding to the third linear prediction coefficient is set; A second reproduction signal is generated by driving with the excitation signal of
A speech and music signal decoding apparatus, wherein a speech and music signal is generated by adding the first reproduction signal and the second reproduction signal. Receiving a first input signal sampled at a first sampling frequency;
Down-sampling the first input signal to a second sampling frequency to generate a second input signal;
Obtaining a first linear prediction coefficient of a second sampling frequency from the second input signal;
A first excitation signal having the second sampling frequency for generating a first reproduction signal corresponding to the second input signal by driving a synthesis filter in which the first linear prediction coefficient is set is obtained. ,
Generating a second reproduction signal by up-sampling the first reproduction signal to the first sampling frequency;
A frequency domain parameter corresponding to a linear prediction coefficient obtained from the first input signal and a frequency corresponding to a second linear prediction coefficient obtained by converting the first linear prediction coefficient into a first sampling frequency by sampling frequency conversion. A third linear prediction coefficient is obtained from the difference from the region parameter,
Obtaining a fourth linear prediction coefficient from a sum of a frequency domain parameter corresponding to the second linear prediction coefficient and a frequency domain parameter corresponding to the third linear prediction coefficient;
A residual signal is generated by driving an inverse filter in which the fourth linear prediction coefficient is set by a difference signal between the first input signal and the second reproduction signal;
Orthogonally transform the residual signal;
A speech music signal characterized by outputting a component corresponding to at least a part of a band of the orthogonally transformed residual signal, the first linear prediction coefficient, the third linear prediction coefficient, and the excitation signal. Encoding method. A first linear prediction coefficient of a second sampling frequency obtained from a second input signal generated by down-sampling the first input signal sampled at a first sampling frequency to a second sampling frequency; The first sampling frequency of the second sampling frequency obtained for driving the synthesis filter in which the first linear prediction coefficient is set to generate the first reproduction signal corresponding to the second input signal. A second linear prediction obtained by sampling frequency conversion of the excitation signal, the frequency domain parameter corresponding to the linear prediction coefficient obtained from the first input signal, and the first linear prediction coefficient into a first sampling frequency A third linear prediction coefficient obtained from the difference from the frequency domain parameter corresponding to the coefficient, and the first reproduction signal as the first sampling frequency. A frequency domain parameter corresponding to the second linear prediction coefficient and the third linear prediction coefficient based on a difference signal between the second reproduction signal generated by up-sampling to a number and the first input signal. Of the residual signal after orthogonal transformation obtained by orthogonal transformation of the residual signal generated by driving the inverse filter in which the fourth linear prediction coefficient obtained from the sum with the corresponding frequency domain parameter is set Receiving a signal including a component corresponding to at least a part of the band;
A first reproduction signal is generated by up-sampling a signal obtained by driving a first linear prediction synthesis filter with the first excitation signal to the first sampling frequency;
Generating a second excitation signal having a first sampling frequency by performing orthogonal inverse transformation on a component corresponding to at least a part of the band of the residual signal after orthogonal transformation;
A second linear prediction synthesis filter in which a signal obtained by adding a frequency domain parameter corresponding to the first linear prediction coefficient and a frequency domain parameter corresponding to the third linear prediction coefficient is set; A second reproduction signal is generated by driving with the excitation signal of
A speech and music signal decoding method, wherein a speech and music signal is generated by adding the first reproduction signal and the second reproduction signal. The component corresponding to at least a part of the band of the orthogonally transformed residual signal, the first linear prediction coefficient of the second sampling frequency, the third linear prediction coefficient of the first sampling frequency, and the second sampling frequency Receiving a signal including a first excitation signal;
A first reproduction signal is generated by up-sampling a signal obtained by driving a first linear prediction synthesis filter with the first excitation signal to the first sampling frequency;
Generating a second excitation signal having a first sampling frequency by performing orthogonal inverse transformation on a component corresponding to at least a part of the band of the residual signal after orthogonal transformation;
A second linear prediction synthesis filter in which a signal obtained by adding a frequency domain parameter corresponding to the first linear prediction coefficient and a frequency domain parameter corresponding to the third linear prediction coefficient is set; A second reproduction signal is generated by driving with the excitation signal of
A speech and music signal decoding method, wherein a speech and music signal is generated by adding the first reproduction signal and the second reproduction signal.

Images